The wave of AI projects reproduces, on an unprecedented scale, the original sin of project management: launching without arbitrating. The problem is not technological. It is strategic.
One figure should alert all executive committees: according to an MIT study published in the summer of 2025, 95% of management projectsGenerative AI carried out in business do not produce any measurable return. The RAND report, for its part, establishes a failure rate greater than 80% for all AI initiatives, double that of traditional IT projects. One might believe there is a problem of technological maturity. This would be a diagnostic error.
Successful projects are not distinguished by superior technology. RAND researchers analyzed the rare successes: in almost all cases, three conditions were met before launch:
- the data domain was already structured,
- the decision-making governance was already clear,
- the use case was formulated in terms of value, not technical prowess.
Not only does AI fail to create strategic alignment, it ruthlessly reveals its absence.
The real subject is not AI. It’s the wallet.
For eighteen months, the CIO are under unprecedented pressure: each business department demands its own AI project, each COMEX wants to display its ambition, and budgets are dispersed into dozens of experiments launched in parallel. MIT calls this phenomenon pilot obsession: a multiplicity of initiatives carried out simultaneously without a scaling strategy. Result: significant costs, exhausted teams, and a project portfolio that no one can control anymore.
This is not a new problem. This is the oldest problem in portfolio management, amplified by fashion. Historically, the majority of project portfolios are not managed: they are suffered. We pile up initiatives without asking the only question that matters: which of these projects really serve the company’s strategy?
AI has not changed this question. It made her scream. When an AI project cost a team’s curiosity, the lack of arbitration was paid for in lost time. When he commits hundreds of thousands of euros and the reputational exposure of the COMEX, he pays not only in cash but above all in dispersion and credibility.
Refereeing doesn’t mean saying yes any faster. It’s knowing how to say no.
The temptation, faced with the wave, is to create an AI committee, a task force, a parallel governance system. This is the opposite of what should be done. An AI project is not a separate category: it is a project, which must enter the same portfolio as the others, be evaluated according to the same strategic value criteria, and compete for the same rare resources. Taking it out of the portfolio on the pretext that it is innovative is precisely what guarantees its failure. RAND specifically identifies this race to the bottom—projects launched for the tool rather than the need—as one of the five root causes of failure.
An aligned portfolio meets three simple requirements that are extremely demanding to maintain. First, each project must be able to state in one sentence the business value it produces and the cost of the inefficiency it eliminates. Then, arbitrations must be made on shared and defensible criteria, not on the balance of power of the moment. Finally, general management must have a clear view of what the portfolio actually produces, not the effort it consumes.
None of this is specific to theartificial intelligence. These are the fundamentals of project portfolio management, those that too many organizations have neglected for years because the absence of management was not expensive enough for us to worry about it.
AI changes this calculus.
It transforms tolerable negligence into a major financial risk. The 5% of companies that get real value from it are not the ones that bought the best technology. These are the ones who already knew how to arbitrage their portfolio before AI arrived. The others discover, abandoned project after abandoned project, that artificial intelligence is a ruthless revealer: it never compensates for the absence of strategy. It adds to the bill.
Before launching the next AI project, only one question deserves to be asked in COMEX. Not “can we do it?” because technology almost always says yes. But “does this project deserve the space and resources that it will take from all the others?” It’s a question of wallet. She always was.
Sources
